mludvig / amazon-textract-parser

Amazon "Textract Results Parser" (trp) module packaged and improved for ease of use.

Amazon Textract Results Parser - textract-trp

Amazon Textract Results Parser or trp module packaged and improved for ease of use.

TL;DR

pip install textract-trp

Requires Python 3.6 or newer.

Usage

import boto3
import trp

textract_client = boto3.client('textract')
results = textract_client.analyze_document(... your file and other params ...)
doc = trp.Document(results)

Now you can examine doc.pages. For example print all the detected on the page:

print(doc.pages[0].text)

Or print out the detected tables in CSV format:

for row in doc.pages[0].tables[0].rows:
    for cell in row.cells:
        print(cell.text.strip(), end=",")
    print()

Or retrieve text from a given position on the page. For that we have to create Bounding Box with the required coordinates relative to the page.

# Coordinates are from top-left corner [0,0] to bottom-right [1,1]
bbox = trp.BoundingBox(width=0.220, height=0.085, left=0.734, top=0.140)
lines = doc.pages[0].getLinesInBoundingBox(bbox)

# Print only the lines contained in the Bounding Box
for line in lines:
    print(line.text)

Refer to the Textract blog post and to amazon-textract-code-samples GitHub repository for more details.

Background

The Amazon blog post about Textract refers to a python module trp.py which used to be quite hard to find. There are many posts on the internet from people looking for the module, often confused by the "other trp module" that's got nothing to do with Textract.

Hence I decided to package and publish the trp.py module from the aws-samples/amazon-textract-code-samples repository. Fortunately its MIT license permits that.

Over time I have made some improvements to the module for ease of use.

Maintainer

Michael Ludvig